Postdoc: High Performance Solution of Three-Dimensional Nonlinear Transient Aeroelastic Problems

9503301法哈特对气动弹性现象的准确预测对安全高性能飞机的设计至关重要。这一预测需要同时求解适当耦合的流体和结构运动方程。因此，气动弹性数值模拟一般都是资源密集型的。我们建议在先进的同质和异质硬件结构上开发用于三维气动弹性问题的并行模拟的高性能方法。更具体地说，我们打算开发用于控制耦合的非线性偏微分方程组的时空并行积分的混合划分方法，并从精度、稳定性、异构性计算、I/O传输、子循环和并行处理等方面考察它们的优点。这项研究是多学科的，包括四个组成部分：计算结构力学、计算流体力学、应用数学和数值分析以及计算机科学和并行处理。前两个部分涉及开发基于有限元和有限体积的离散数学模型，用于将航空航天结构表示为梁、板和壳的组件，并对它们周围的高速流动进行建模。第三部分设计交错和高效的时间积分器，以推进控制耦合半离散方程的求解。最后一个研究部分旨在利用同构和异质并行系统的能力来显著减少这些计算密集型工程应用程序的总体模拟时间。

9503301 Farhat The accurate prediction of aeroelastic phenomena is essential to the design of safe high performance aircraft. This prediction requires solving simultaneously the appropriately coupled fluid and structural equations of motion. Therefore, numerical aeroelastic simulations are in general resource intensive. We propose to develop high performance methodologies for the parallel simulation of three-dimensional transient aeroelastic problems on advanced homogenous and heterogeneous hardware architectures. More specifically, we intend to develop mixed partitioning procedures for the parallel space and time integration of the governing coupled nonlinear partial differential equations, and investigate their merits in terms of accuracy, stability, heterogeneous computing, I/O transfers, subcycling, and parallel processing. This research is multidisciplinary and includes four components: computational structural mechanics, computational fluid dynamics, applied mathematics and numerical analysis and, computer sciences and parallel processing. The first two components involve the development of discrete mathematical models based on finite elements and finite volumes for representing aerospace structures as assemblies of beams, plates, and shells, and modeling high speed flows around them. The third component addresses the design of staggered and efficient time-integrators for advancing the solution of the governing coupled semi-discrete equations. The last research component aims at exploiting the capabilities of homogeneous and heterogeneous parallel systems to significantly reduce the overall simulation time of these very compute-intensive engineering applications.